BHASHINI and IISc Bengaluru Build Four Text-to-Speech Models for the Low-Resource Sourashtra Language
The Digital India BHASHINI Division (DIBD) under MeitY, working with IISc Bengaluru, developed four Text-to-Speech models for Sourashtra, a low-resource Indo-Aryan language spoken mainly in Tamil Nadu. Built on the VITS architecture, the models generate speech at 22,050 Hz and offer male and female voices in both the Tamil and Sourashtra scripts. Training used a corpus from IISc's SPIRE Lab holding over 28 hours of recordings, with about 10,000 male and 11,400 female utterances. The models are published on BHASHINI's AIKosh platform, extending speech technology to a language with almost no digital footprint and advancing digital inclusion.
BHASHINI, the National Language Translation Mission, was launched in 2022 under MeitY to build open AI-based language technology for Indian languages | AIKosh is India's national repository of AI datasets, models and toolkits, launched in 2025 | The Eighth Schedule of the Constitution lists 22 scheduled languages; Sourashtra is not among them